computer components & hardware - Graphics performance and 64 bit OS

*sigh* One more time, then I give up:

----
FROM: http://www.osnews.com/story.php?news_id=5768

"So while the tests I ran were on only a few applications and in limited
ways, the results seem to show that indeed 64-bits do generally run
slower. However, there are there are a few issues to consider."

----
FROM: http://arstechnica.com/cpu/03q1/x86-64/x86-64-1.html

"...the only kind of performance increase that you can expect from a
straight 64-bit port is whatever additional performance you get from
having more memory available. As I said earlier, 64-bitness, by itself,
doesn't really improve performance for anything but the rare 64-bit
integer application. In the case of x86-64, it's the added registers and
other changes that actually account for better performance on normal
apps like games."
----

I'm not exactly just pulling random opinions out of my ass here. I have
worked for several years in the software industry and have designed and
run code for 8-, 16-, 32-, and 64-bit processors. The only statement
I'm trying to make, which I've pointed out with each post to this
thread, is that "64-bit does not mean inherently faster performance."

You even seem to support my statement, with your argument of "... then
why does 64-bit Linux run so much faster than 32-bit Linux on the same
hardware?"

The answer is obvious and was addressed in my previous post with the
second statement: "*If* the code has been recompiled to take advantage
of the new 64-bit features, it will perform better; otherwise, not."
The 32-bit Linux code running on Athlon64 architecture cannot "see" the
extra registers, and cannot therefore use them in anyway. Only by
recompiling the code with the 64-bit flag can you 'force' the code to
recognize the extra hardware and utilize it.

I fail to see the "variance" in my claims and I am sorry that I cannot
be anymore clear or straightforward with "my nice opinions."

J. Clarke wrote:

cK-Gunslinger wrote:

That's *exactly* what I just said. *If* the code has been recompiled to
take advantage of the new 64-bit features, it will perform better;
otherwise, not.

A 64-bit CPU gives you the ability to use 64-bit memory address,
allowing you to have a *lot* more addressable memory that 32-bit (which
are 'limited' to ~4GB). Also, the registers are 64-bit, allowing you do
perform 64-bit calculations in a single step, as opposed to doing them
twice with 2 32-bit numbers. That's it. That's all that's required for
a CPU to be '64-bit'. If your applications use a lot of 64-bit
number/integers (most do not), then you might see some speedup,
Otherwise, not. Simple.

64-bit = 'more memory, larger native numbers'
64-bit != 'inherently faster performance'

But like I said before, all the new 64-bit CPUs include numerous other
architectural enhancements that will outperform current 32-bit CPUs, so
your apps *will probably* perform better on an Opteron than your current
p4, but *not* because the Opteron is 64-bit, but rather in spite of it.

That's a nice opinion, now, do you have any examples using real-world
processors and real-world code to support it?

What you claim to be saying and what you are actually saying seem to be more
than a little at variance. If the advantages you cite are the only ones
that a 64-bit processor provides then why does 64-bit Linux run so much
faster than 32-bit Linux on the same hardware?

J. Clarke wrote:

cK-Gunslinger wrote:

All things being equal, a 64-bit machine (CPU + OS) will typically be
*slower* than a 32-bit machine. Now, all things are *never* equal and
the current line-up of 64-bit CPUs (AMD64, Opteron, etc) have many
hardware enhancements (extra registers, larger caches, faster memory
interconnects, etc) that will increase overall performance.

That being said, if you own a 64-bit CPU and are running in 32-bit mode,
will you get better performance simply by installing WindowsXP-64 and
some 64-bit drivers? Possibly, but only if the OS/drivers have been
tweaked to take advantage of the CPU/chipset's new features. The
general "64-bit-ness" of the software does not make anything faster over
32-bit.

I'd be interested in knowing on what you base this conclusion.

FWIW, with the Opterons Linux shows a marked performance improvement when
recompiled as 64-bit code.